Individual site estimates 1.#QNB - 1.#QNB

[lstuemke]

Any guidance would be greatly appreciated for this question.

For my data I am looking at two species that have very low detection probabilities, with 10 repeated surveys in 20 separate study units. I've run single-season analysis with covariates. When modeling the covariates as one per model some models generate individual site estimates and other models generate the following:

Individual Site estimates of <Psi>
Site Survey Psi Std.err 95% conf. interval
1 site 1 1-1: 1.0000 0.0000 1.0000 - 1.0000
2 site 2 1-1: 1.0000 0.0000 1.0000 - 1.0000
3 site 3 1-1: 1.0000 -1.#IND 1.#QNB - 1.#QNB
4 site 4 1-1: 1.0000 -1.#IND 1.#QNB - 1.#QNB
5 site 5 1-1: 1.0000 -1.#IND 1.#QNB - 1.#QNB
6 site 6 1-1: 1.0000 -1.#IND 1.#QNB - 1.#QNB
7 site 7 1-1: 1.0000 -1.#IND 1.#QNB - 1.#QNB
8 site 8 1-1: 1.0000 -1.#IND 1.#QNB - 1.#QNB
9 site 9 1-1: 1.0000 0.0000 1.0000 - 1.0000
10 site 10 1-1: 1.0000 0.0000 1.0000 - 1.0000
11 site 11 1-1: 1.0000 -1.#IND 1.#QNB - 1.#QNB
12 site 12 1-1: 1.0000 -1.#IND 1.#QNB - 1.#QNB
13 site 13 1-1: 1.0000 0.0000 0.6410 - 1.0000
14 site 14 1-1: 1.0000 0.0000 1.0000 - 1.0000
15 site 15 1-1: 1.0000 0.0000 1.0000 - 1.0000
16 site 16 1-1: 1.0000 0.0000 1.0000 - 1.0000
17 site 17 1-1: 1.0000 0.0000 1.0000 - 1.0000
18 site 18 1-1: 1.0000 -1.#IND 1.#QNB - 1.#QNB
19 site 19 1-1: 0.0000 0.0000 0.0000 - 0.4645
20 site 20 1-1: 1.0000 -1.#IND 1.#QNB - 1.#QNB

Is this a result of minimal detection histories and low detection probabilies?
How should I handle these models?
Thoughts?

[jhines]

When you see results like that, it is usually due to insufficient data or trying to estimate parameters as a function of too many covariates. With sparse data, you often get estimates like those. If I had to guess, I'd say that all sites with one value of the covariate were occupied (or had detections) and all sites with the other value of the covariate were unoccupied. With only 20 sites and very low detection probabilities, you're probably not going to be able to run a model with covariates as you need a mix of sites with detections and non-detections for both/all values of the covariate in order to get estimates.

Jim

[lstuemke]

Thanks Jim.